Physical internet-enabled hyperconnected distribution assessment

(1)

Physical Internet-Enabled

Hyperconnected Distribution Assessment

Thèse

Helia Sohrabi

Doctorat en sciences de l’administration

Philosophie doctor (Ph.D.)

Québec, Canada

(2)

(3)

Résumé

L'Internet Physique (IP) est une initiative qui identifie plusieurs symptômes d'inefficacité et non-durabilité des systèmes logistiques et les traite en proposant un nouveau paradigme appelé logistique hyperconnectée. Semblable à l'Internet Digital, qui relie des milliers de réseaux d'ordinateurs personnels et locaux, IP permettra de relier les systèmes logistiques fragmentés actuels. Le but principal étant d'améliorer la performance des systèmes logistiques des points de vue économique, environnemental et social.

Se concentrant spécifiquement sur les systèmes de distribution, cette thèse remet en question l'ordre de magnitude du gain de performances en exploitant la distribution hyperconnectée habilitée par IP. Elle concerne également la caractérisation de la planification de la distribution hyperconnectée. Pour répondre à la première question, une approche de la recherche exploratoire basée sur la modélisation de l'optimisation est appliquée, où les systèmes de distribution actuels et potentiels sont modélisés. Ensuite, un ensemble d'échantillons d'affaires réalistes sont créé, et leurs performances économique et environnementale sont évaluées en ciblant de multiples performances sociales. Un cadre conceptuel de planification, incluant la modélisation mathématique est proposé pour l’aide à la prise de décision dans des systèmes de distribution hyperconnectée.

Partant des résultats obtenus par notre étude, nous avons démontré qu’un gain substantiel peut être obtenu en migrant vers la distribution hyperconnectée. Nous avons également démontré que l'ampleur du gain varie en fonction des caractéristiques des activités et des performances sociales ciblées.

Puisque l'Internet physique est un sujet nouveau, le Chapitre 1 présente brièvement l’IP et hyper connectivité. Le Chapitre 2 discute les fondements, l'objectif et la méthodologie de la recherche. Les défis relevés au cours de cette recherche sont décrits et le type de contributions visés est mis en évidence. Le Chapitre 3 présente les modèles d'optimisation. Influencés par les caractéristiques des systèmes de distribution actuels et potentiels, trois modèles fondés sur le système de distribution sont développés. Chapitre 4 traite la caractérisation des échantillons d’affaires ainsi que la modélisation et le calibrage des paramètres employés dans les modèles. Les résultats de la recherche exploratoire sont présentés au Chapitre 5. Le Chapitre 6 décrit le cadre conceptuel de planification de la distribution hyperconnectée. Le chapitre 7 résume le contenu de la thèse et met en évidence les contributions principales. En outre, il identifie les limites de la recherche et les avenues potentielles de recherches futures.

(4)

(5)

Abstract

The Physical Internet (PI) is an initiative that identifies several symptoms of logistics systems unsustainability and inefficiency and tackles them by proposing a novel paradigm called Hyperconnected Logistics. Similar to the Digital Internet, which connects thousands of personal and local computer networks, PI will connect the fragmented logistics systems of today. The main purpose is to enhance the performance of logistics systems from economic, environmental and social perspectives.

Focusing specifically on the distribution system, this thesis questions the order of magnitude of the performance gain by exploiting the PI-enabled hyperconnected distribution. It is also concerned by the characterization of the hyperconnected distribution planning. To address the first question, an exploratory research approach based on optimization modeling is applied; first, the current and prospective distribution systems are modeled. Then, a set of realistic business samples are created, and their economic and environmental performance by targeting multiple social performances are assessed. A conceptual planning framework is proposed to support the decision making in the hyperconnected distribution system.

Based on the results obtained by our investigation, it can be argued that a substantial gain can be achieved by shifting toward Hyperconnected Distribution. It is also revealed that the magnitude of the gain varies by business characteristics and the targeted social performance.

Since the Physical Internet is a novel topic, chapter 1 briefly introduces PI and Hyperconnected Logistics. Chapter 2 discusses the research foundations, goal and methodology. It also describes the challenges of conducting this research and highlights the type of contributions aimed for. Chapter 3 presents the optimization models including a core distribution network design modeling approach. Influenced by the characteristics of the current and prospective distribution systems, three distribution system-driven models are developed. Chapter 4 engages with the characterization of the business samples, the modeling and calibration of the parameter that are employed in the models. The exploratory investigation results are presented in Chapter 5. Chapter 6 describes the hyperconnected distribution planning framework. Chapter 7 summarizes the content of the thesis and highlights the main contributions. Moreover, it identifies the research limitations and potential future research avenues.

(6)

(7)

List of tables

Table 2-1. The evolution of the distribution era... 20

Table 2-2. Summary of the main research blocks ... 26

Table 3-1. The modeling aspects incorporated in a selected set of single-period MIP distribution network design models ... 32

Table 4-1. Developed businesses cases characteristics summary ... 62

Table 4-2. The target service level in terms of maximum tolerated outbound distance by truck for three market zone classes ... 63

Table 4-3. The average annual costs associated with DC opening and operations ... 68

Table 4-4. Geographical impact factor applied to the area-based DC cost ... 70

Table 4-5. Discounted transportation cost as piecewise linear function of quantity ... 72

Table 4-6. Full truck load and less than truck load unitary transportation cost as stepwise function of distance (m(nn)) ... 74

Table 4-7. Full truck load and less than truck load unitary transportation cost as stepwise function of distance ... 75

Table 5-1. Summary of the Key Performance Indicators characteristics ... 80

Table 5-2. Transportation-induced energy consumption and GHG emission generated (from CLECAT guide, 2012) ... 82

Table 5-3. Environmental impact parameters associated with each shipment option ... 85

Table 5-4. Environmental impact measures associated with distribution center building and involved operations ... 88

Table 5-5. Distribution center-induced energy consumption and GHG emission generated for conventional and PI-enabled open DCs ... 89

Table 5-6. Characteristics of multiple distribution system-driven MIP problems solved in the investigation ... 90

Table 5-7. Total dedicated distribution cost responding to each service level in [M$/year] ... 91

Table 5-8. Economic total distribution cost impact of upgrading service level in dedicated distribution system



_{ }d__ ... 92

(12)

Table 5-9. Number of DCs opened and total annual DC opening and warehousing cost (briefly DC cost) responding to each service level ... 94 Table 5-10. Additional number of DCs and opening and warehousing cost required to respond to higher market service level



_{ }d__ ... 94 Table 5-11.Transportation cost performance and impact by responding to higher service level ... 95 Table 5-12. Total energy consumption responding to each service level in [TJ/year] ... 98 Table 5-13. Total GHG production responding to each service level in [MKgCO2.equivalent/year]

... 98 Table 5-14. Energy consumption impact of upgrading service level in dedicated distribution system ... 98 Table 5-15. GHG production impact of upgrading service level in dedicated distribution system .. 99 Table 5-16. Total collaborative distribution cost responding to each service level [M$/year] ... 100

Table 5-17. Intra-collaborative distribution system cross-service level economic performance



cjn ... 100 Table 5-18. Number of DCs opened at each echelon and total annual DC opening and warehousing cost responding to each service level ... 102 Table 5-19. Additional DC opening and warehousing cost required to respond to higher market service level ... 102 Table 5-20. Transportation cost performance and impact by responding to higher service level ... 103 Table 5-21. Energy consumption of the collaborative distribution system responding to each service level ... 106 Table 5-22. GHG emission production of the collaborative distribution system responding to each service level ... 106 Table 5-23. Energy consumption impact of upgrading service level in collaborative distribution system c

jn



 ... 106

Table 5-24. GHG emission production impact of upgrading service level in collaborative

distribution system



_c_jn ... 107 Table 5-25. Total hyperconnected distribution cost responding to each service level in [M$/year] ... 108

(13)

Table 5-26. Intra-hyperconnected distribution cost cross-service level (



dj) ... 108 Table 5-27. Energy consumption of the hyperconnected distribution system responding to each service level ... 110 Table 5-28. GHG emission production of the hyperconnected distribution system responding to each service level ... 110 Table 5-29. Energy consumption impact of upgrading service level in hyperconnected distribution system



hj   ... 111

Table 5-30. GHG emission production impact of upgrading service level in hyperconnected

distribution system



hj  ... 111

(14)

(15)

List of figures

Figure 1-1. The first PI-container design (left) and product (right) by MODULUSHCA project ... 3

Figure 1-2. Schematic representation of the composition-decomposition of PI-containers based on the Physical Internet manifesto (Montreuil, 2012) ... 3

Figure 1-3. Physical Internet enabled Logistics Web and its key constituents (adapted from Montreuil et al., 2013) ... 4

Figure 1-4. Illustrative road-based hub network of Physical Internet-enabled Mobility Web ... 6

Figure 1-5. Conventional DC and PI-enabled open DC contrast ... 7

Figure 2-1. The distribution system as a dynamic system ... 13

Figure 2-2. The distribution system categorization according to the dynamic system entities ... 17

Figure 2-3. Schematic contrast of dedicated, collaborative and hyperconnected distribution systems ... 19

Figure 2-4. The research methodology through four major contributions ... 23

Figure 3-1. Optimization phase within the research approach ... 28

Figure 3-2. An anticipation-based modeling approach (based on Klibi et al., 2015) ... 35

Figure 3-3. The distribution network structure ... 36

Figure 3-4. Schematic contrast of shipment options defined here ... 38

Figure 3-5. Discounted transportation flow as a piecewise linear function of quantity ... 47

Figure 3-6. Unitary transportation cost as a stepwise function of distance ... 48

Figure 3-7. Schematic demonstration of a 2-business collaborative distribution web ... 50

Figure 3-8. Schematic demonstration of a hyperconnected distribution web ... 56

Figure 4-1. The business case characterization and key parameter modeling highlighted in the research approach ... 60

Figure 4-2. Location of the complete set of market zones and potential distribution centers ... 61

Figure 4-3. Low throughput business typical business territory, demand distribution and market classification ... 63

Figure 4-4. Average throughput business sample territory, demand distribution and market classification ... 64

Figure 4-5. High throughput business sample territory, demand distribution and market classification ... 64

(16)

Figure 4-6. Typical layout considered here for potential DCs ... 65 Figure 4-7. The scale curve applied to area-based DC cost to capture the impact of the economies of scale ... 66 Figure 4-8. Validation of the estimated discrete values of total DC cost with the impact of

economies of scale ... 69 Figure 4-9. Function of the discounted flow because of economies of scale ... 71 Figure 5-1. Performance assessment and potential gain investigation highlighted in the research approach ... 78 Figure 5-2. The environmental impact evaluation approach ... 81 Figure 5-3. Dedicated distribution network of the typical average throughput business responding to the top service level ... 97 Figure 5-4. Dedicated distribution network of the typical average throughput business responding to the basic service level ... 97 Figure 5-5. Collaborative distribution web of the typical average throughput business responding to the top service level ... 104 Figure 5-6. Collaborative distribution web of the typical average throughput business responding to the basic service level ... 104 Figure 5-7. Hyperconnected distribution web of the average throughput typical business responding to the top service level ... 109 Figure 5-8. Hyperconnected distribution web of the average throughput typical business responding to the basic service level ... 109 Figure 5-9. Collective economic performance gain in terms of total distribution cost as a function of service level ... 112 Figure 5-10. High-throughput business sample economic performance gain in terms of total

distribution cost as a function of service level ... 112 Figure 5-11. Average-throughput business sample economic performance gain in terms of total distribution cost as a function of service level ... 113 Figure 5-12. Low-throughput business sample economic performance gain in terms of total

distribution cost as a function of service level ... 113 Figure 5-13. Collective economic performance gain in terms of total DC opening and warehousing cost as a function of service level ... 114

(17)

Figure 5-14. Collective economic performance gain in terms of total transportation cost as a

function of service level ... 115

Figure 5-15. Collective environmental performance gain in terms of total GHG emission production as a function of service level ... 116

Figure 5-16. Collective environmental performance gain in terms of total energy consumption as a function of service level ... 116

Figure 6-1. Hyperconnected distribution planning framework underlined in the research approach ... 120

Figure 6-2. The interlaced levels of Hyperconnected Distribution Planning framework ... 122

Figure 6-3. Modeling approach (inspired by Klibi et al, 2012b) ... 125

Figure 6-4. The two-stage decision making process along the planning horizon ... 128

(18)

(19)

To my parents, my sister and grandmother To Nicolas And the everlasting memory of Pegah

(20)

(21)

Acknowledgment

Every page of this thesis reminds me of encouraging words and support of those who believed in me, loved me and were there for me, so I have made it through. Here, I share my true admiration and deepest appreciation with them.

I have been extremely fortunate to be supervised by Professor Benoit Montreuil. He offered me so much beyond what could be found in any book; his visions, trans-domain knowledge, research experience and teaching advice. I am sincerely thankful for his generous financial support, for pushing me to get out of my shell and for letting me grow not just as a researcher but also as a person.

During the last four years, my research has been enriched by co-direction of Professor Walid Klibi. He invested enormous amount of time and effort to advise me particularly with the technical aspects of my research. I would like to thank him for his financial support during my internship at BEM, for his availability and patience and lastly, for believing in my abilities at the times I myself did not.

Words fall short to express my gratitude for my parents. They have supported me throughout my life and particularly to come to Canada. My Ph.D. endeavor was as much their dream as was mine. Their sacrifice and encouragement gave me the chance to chase my dreams and for that chance, I am forever grateful. My sister, Hengameh, and my grandmother, Mamani, send me the words of encouragement when I needed them the most; they have never been far from me while being at the other side of the world. I am truly lucky for the heartwarming love of my boyfriend, Nicolas, whose sweet care and understanding helped me to face difficulties. I have been also inspired by his scientific experience and rigorous research conduction. Looking up to him encouraged me to strive for greater achievements.

Many kind and caring people have crossed my Ph.D. path; they showed me around when I arrived to Canada, listened to me when I was stressed over my studies and made me feel like home when I was homesick. Although it is not possible for me to thank them all in here, I would like to name a few; Niloofar Kazemi, Shirin Edarehchi, Matthew Guillemette, Parisa Yousefi, Daniel Kreeft, Lorena Ruelas, Michel and Marie Venkovic.

At Laval University, I would like to thank Professor Jacques Renaud, the vice dean of research in the Business Administration faculty, and Professor Bernard Lamond, the head of Operations and Decision Systems department for the teaching opportunities and financial aids provided by them. Special thanks to Mr. Laurent Duchesne from CalculQuebec organization at Laval University for

(22)

facilitating the computations of my research on Colosse supercomputer. At CIRRELT research center, my genuine regards go to Edith Brotherton, Caroline Clautier, Louise Doyon and Pierre Marchand for their kind advices and help in various matters. Furthermore, the technical guidance of my colleagues Mr. Alexis Roy, CIRRELT network manager and Mr. François Barriault, CIRRELT research professional are truly appreciated. I would like to thank other Ph.D. students and researchers in the group, Vincent Augusto, Maryam Darvish, Driss Hakimi and Salma Naccache with whom I shared many interesting conversations.

Besides those I have mentioned here, I would like to extend my deepest gratitude to my teachers in Iran, particularly Mrs. Machinchi, Mrs. Safar pour, Mrs. Elmi and Mrs. Sadeh and all of those who helped me, directly or indirectly, to be who I am today.

(23)

1

Chapter 1

(24)

1.1 Fundamental definitions

The Physical Internet is defined to be a “hyperconnected global logistics system enabling seamless

asset sharing and flow consolidation” (Montreuil, 2015). Montreuil (2015) defines a hyperconnected

system as one whose components are intensely interconnected on multiple layers, ultimately anytime

and anywhere.

Let us first discuss the definition of the term, interconnection. It means the purposeful connection of more than one entities without altering their independency (based on Cambridge dictionary and Kahn and Cerf, 1999). The interconnection can be created for various purposes such as information transmission, financial transactions, collaboration in operations or physical asset sharing.

A well-known instance of interconnected systems is the digital Internet. The Internet is defined as a global information system formed by the interconnection of numerous independent computers that is federated into a seamless whole without changing any of the underlying networks (Kahn and Cerf, 1999). Our personal computers, same as hundreds of thousands of local and global computer networks interconnect with each other by transmitting data through Internet protocols.

When the level of interconnectivity intensifies by interconnection of entities on multiple layers, such as digital, physical, operational, business and legal to name a few, it is called hyperconnectivity (based on Montreuil, 2015). The Physical Internet aims to create hyperconnectivity by enabling intensive interconnection among currently dedicated and privately operated logistics systems without altering their independency. A large part of the potential to develop such hyperconnection already exists. Most of the current logistical facilities are owned or leased by an individual firm or a group of collaborative firms (Montreuil et al, 2013). Only in United States, there are in order of five hundred thousand distribution centers and warehousing facilities (Montreuil, 2011) and more than 10 million trucks (US Department of Transportation-Bureau of transportation studies, 2012). The average utilization rate of DCs and warehouses is reported to be on average 70% (Ecklund, 2010) and commercial trucks, almost 60% (McKinnon et al., 2010; Sarraj et al., 2014). These statistics indicate an existing opportunity to interconnect businesses by sharing assets and consolidating transportation flows (physical and operational interconnectivities).

In the Physical Internet, products are embedded in standard and modular size PI-containers equipped with RFID technologies to carry and transmit information. The size of PI-containers ranges from box size to cargo container size in a Lego shape that enables consolidation of tens of PI-containers to enhance the filling rate of the transportation vehicles, as depicted in Figure 1-1 and Figure 1-2.

(25)

Figure 1-1. The first PI-container design (left) and product (right) by MODULUSHCA project (Landschützer et al., 2014)

Figure 1-2. Schematic representation of the composition-decomposition of PI-containers based on the Physical Internet manifesto (Montreuil, 2012)

The standard size of PI-containers facilitates their handling and movement with standard material handling equipment Papers by Montreuil et al. (2012 and 2013) and Ballot et al. (2013) provide in depth coverage of PI-containers application and their impact on material handling technology and unimodal and multimodal PI-facilities.

The PI digital interconnectivity can be achieved by the exchange of information, such as the origin and destination of the product stored inside the container between users, service providers and operators. The digital interconnectivity ensures a seamless exchange of meaningful information and

(26)

fact-based decision-making and action between PI-users. Moreover, it enables their distributed controlling and planning, for example for the purpose of efficient routing and consolidation (Sarraj et al., 2014).

Similar to the digital Internet and its web sites and applications built upon the digital Internet, businesses can exploit the Physical Internet through PI-enabled Logistics Web (Montreuil et al., 2013). The term “web” is used to represent a network of networks (Hakimi et al., 2009). The Logistics Web involves the network of open facilities, technologies and services. The openness feature refers to the characteristics of PI components (such as facilities and technologies) to be available for the use of PI-certified users other than their owner. As summarized in Figure 1-3, the Logistics Web embraces five sub-webs including Realization Web, Service Web, Supply Web, Mobility Web and Distribution Web.

Figure 1-3. Physical Internet enabled Logistics Web and its key constituents (adapted from Montreuil et al., 2013) PI-enabled Logistics Web Realization Web Sup ply_W eb _Service W eb

(27)

In accordance to the subject of this thesis, more insights about the Mobility Web (MW) and the Distribution Web (DW) are provided in the following section.

1.2 PI-enabled Mobility Web

The Mobility Web is enabled by the Physical Internet for the purpose of efficiently and seamlessly moving and handling goods encapsulated in PI-containers between their individual sources and destinations. MW is built upon the hyperconnection of unimodal and multimodal transportation infrastructure, transporters, vehicles, drivers, hubs and terminals (Montreuil et al., 2013). For more insights, let us contrast in this section the conventional truck mobility networks and the PI-enabled truck mobility web.

The conventional truck mobility network is dedicated to a single firm or a group of partnering firms. The dedication can be in the form of infrastructure ownership, lease or outsourcing. In contrast, the PI-enabled Mobility Web is open to all certified PI-users. For instance, the available space in an open truck can be exploited by multiple PI-users for the entire or a part of the truck’s shipment path. The trailer of an open truck is unloaded completely or partially in a PI-hub, and then off loaded containers will travel the remaining of the path in other trucks. The decisions regarding the selection of the hub(s) to load/unload PI-containers are controlled and implemented in a distributed manner (Sarraj et al., 2014). Influenced by such an open pooling, the truck filling rate is subject to increase (Ballot et al. 2014). Accordingly, higher filling rates are to result in lower transportation cost, fuel consumption, and GHG emission production.

The Physical Internet intends to improve the quality of life of truck drivers. Montreuil (2011) refers to truck drivers as “modern cowboys” whose lives are affected the most by long distance travels and absences from home. Hence, the Physical Internet implements a geographically extended web of open transportation hubs and terminals that are strategically located so that truck drivers can potentially work in short-distance from their hometown. For instance, selecting the location of PI-hubs at the junction of states’ border and interstates highway/railroad, gives the drivers the chance to operate inside a single state or within a limited region. Figure 1-4 envisions an open PI-hub backbone network for road transportation mode in US and Canada. The stars indicate attractive potential open hub zones that are strategically located at the intersection of major highways and the border of US States/Canadian provinces. More advanced versions of the backbone network should include many more hub zones to take into account other transportation modes such as air, rail and maritime and the multi modal hubs.

(28)

Figure 1-4. Illustrative road-based hub network of Physical Internet-enabled Mobility Web

It should be noted that, businesses may create a combination of dedicated and hyperconnected mobility. Extremes have a business relying only on its own dedicated network and a business relying only on the Mobility Web. Many businesses will take a hybrid position, for example selecting to keep steady full-truckload shipments for their inbound transportation and relying on the MW for the rest.

In chapter 2 and 3, we discuss the benefits of the truck conventional mobility in terms of savings through economies of scale and distance.

1.3 PI-enabled Distribution Web

The Distribution Web is enabled by the Physical Internet for the purpose of efficient and seamless storage, deployment and cross docking of goods encapsulated in PI-containers. DW is built upon the hyperconnection of users with open distribution centers and warehouses. Similar to our discussion about Mobility Web, conventional distribution networks are dedicated to a single or a group of individual businesses, while capacities and technologies of an open DC can be exploited by any PI-users. Figure 1-5 contrasts a distribution center operated by a single company (part a), operated by a group of companies (part b) and open to any PI-user (part c). It also shows the similar concept for conventional and open truck.

(29)

a) Conventional DC operated by a single business

b) Conventional DC operated by a group of partnering businesses

c) PI-enabled open DC exploited by any PI-user Figure 1-5. Conventional DC and PI-enabled open DC contrast

(30)

Suppose that a PI-user connects to the database of the Distribution Web and searches for available open DCs in the cloud. According to Montreuil et al. (2013), a plausible scenario is that a set of openly available DCs would appear in the search result indicating the offered capacity by open DC, availability duration and price. A PI-user interested in an open DC can request to exploit the available capacity partially or entirely. After the agreement of both parties (the open DC owner and user), the capacity requested by the PI-user will be reserved in the open DC for the duration of their agreement.

The open DC can store, pick, move and consolidate the PI-containers belonging to all of its users. High throughput and application of both modular material handling technologies and standard PI-containers enables a high level of efficiency in an open DC. Hence, it is expected that capacity utilization rate of open DCs becomes higher than the conventional DCs and the unitary cost to exploit an open DC would be much lower than conventional DCs.

A real world instance of a nearly open DC can be found in the ES3 distribution facility in York, PA1_.

This is the biggest automated grocery warehouse in USA. It stores around 2000 different items, and its throughput arrives in more than 100,000 boxes per day or 1,300 pallets per day. The complex is 140,000 m2_{at and can potentially store 400,000 pallets. ES3 has established a successful collaborative}

distribution model, storing and consolidating products of 60 manufacturers, creating seamless operations, providing real-time information and almost 30% reduction in total cost of getting products from manufacturer to retailer for both parties (Hambleton and Mannix, 2014).

Handling of the slow moving products is fully automated and the remaining is handled both automatically and manually (based on the statistics provided by ES3 executives and the supplier of their automation system; SSI SCHAEFER2_{). According to Hambleton and Mannix (2014), ES3’s}

automated material handling system can result in fewer operations, which translate in lower required labor, energy savings (almost 40%) and lower accident rates.

1.4 Conclusion

In this chapter a brief review of Physical Internet principles, particularly related to the distribution and transportation is provided. The full presentation and description of Physical Internet is beyond the scope of this thesis. The reader is invited to consult pioneering PI documents such as Montreuil

(31)

(2011), Montreuil et al. (2013) and Ballot et al, 2014. The website of the research initiative is also regularly updated by the recent progress and findings (www.physicalinternetinitiative.org).

The next chapter defines the research foundations, presents the research questions and goals, and describes the research approach.

(32)

(33)

2

Chapter 2

(34)

2.1 Introduction

Physical distribution is about channeling the products of a business to its customers. The decisions and operations involved in distribution strongly impact the economy and environment. The costs related to distribution, mainly transportation, inventory holding cost and storage represent almost 90% of logistics operations costs (Rushton, 2010). Moreover, the transportation sector largely contributes to the Greenhouse Gas emission; almost 26% of the total USA emission in 2013 and 24% of Canada’s emission in 2011 is accounted for transportation activities (based on US Environmental Protection Agency and Environment Canada). On the other hand, distribution and transportation operations strongly impact the customer service. One of the key business competitive dimensions is customer satisfaction, which results in revenue increase. Based on the discussion above, the distribution is one of the most potentially value adding areas of logistics operations.

Despite its important role, several studies provide evidence of distribution inefficiency and unsustainability. For instance, Montreuil (2011) criticizes the centralization of storage and distribution operations to a few large facilities as it negatively impacts the response time and customer service quality. Frankle (2006) reports that 8.2% of shoppers fail to find product(s) on-shelf and retailers suffer 3.1% net lost sales while huge inventory valued at 1.1 trillion$ have been available. Significant portions of consumer products that are made never reach the right market on time, ending up unsold and unused (Montreuil, 2011). These findings can be simply summarized to the right product is not at the right place at the right time at the right cost in the right quantity!

Moreover, capacities and technologies within the distribution and storage facilities are being poorly used (Montreuil, 2011); Ecklund (2010) indicates the capacity utilization rate of warehouse to be on average 70%. The capacity of transportation fleet is also inefficiently used; almost one-third of kilometers traveled by freight transport vehicles are run empty (McKinnon, 2000) and the mean load factor of road transport is approximately 60% (Sarraj et al., 2014; McKinnon et al., 2010). From social performance perspective, Montreuil (2011) highlights the truck drivers’ quality of life; he refers to them as “modern cowboys” who are often far away from home for long duration.

This Ph.D. thesis aims to assess a novel distribution system, called Physical Internet-enabled hyperconnected distribution that is claimed to provide means to enhance the efficiency and sustainability of distribution systems by an order of magnitude (Montreuil, 2011; Montreuil et al., 2013).

(35)

Through the first part of this chapter, the evolution of distribution systems to the date is reviewed and our categorization of the existing and prospective distribution systems is presented. This categorization serves as the foundation of our research methodology. The second part of this chapter describes the research question and goal and details our research approach.

2.2 Distribution system definition

Inspired by system dynamics (Forrester, 1958; Forrester, 1992), we define the distribution system of an individual business as the logical and physical manifestation of all the strategies, decisions and operations intended for deploying, storing, handling and delivering products. Noteworthy, here by using the term business, we refer to an organization that acquires products (whether by producing or purchasing them) and is concerned by their physical distribution. Figure 2-1 represents the distribution system as here defined.

Figure 2-1. The distribution system as a dynamic system

The entities of the distribution system include the stakeholder(s), the physical topology, the logical topology, the operations and the performance. The term stakeholder refers to the decision makers and managers whose opinion, decisions and actions impacts the entire system. They can be the owners or

(36)

investors of a business. Stakeholder(s) define a set of rules and policies for the distribution-related decision making that is called logical topology. Following such rules, they form the physical topology by determining the location and mission of various facilities including the distribution center(s) and warehouse(s). The logical and physical topologies of the system prepare the foundation for the distribution operations. The resulting distribution performance provides a feedback, whether in forms of return on investment, distribution-related costs, customer satisfaction or environmental performance, to the stakeholder(s).

2.3 Distribution system evolution

Before 1960, distribution activities were unplanned and unformulated (Rushton, 2010). At this period, management did not view the distribution as an integrated task; thus, it was carried out with a series of fragmented, uncoordinated movement of goods and information (Rushton, 2010). The logistics at this period is called atomistic by Montreuil (2015).

The concept of physical distribution as a valid area for managerial involvement was developed in the 1960s and early 1970s. Subsequently, the 1980s was the growth of the Third-Party Logistics Providers or 3PLs (Rushton, 2010). By outsourcing the distribution operations to 3PLs, firms can benefit from reduction in asset investment, labor and equipment maintenance cost, focusing on core competencies and exploiting the external expertise. The globalization trend and increased importance of partnership in staying competitive distinguished the 3PL concept as a differentiator in company’s competitive business (Papadopoulou and Macbeth, 1998). Besides its advantages, outsourcing reduces the control of an organization over logistics functions and impacts in-house capability and customer contact (Selviaridis and Spring, 2007).

In late 1980s and early 1990s, the interrelation between distribution-related activities such as transportation, storage, material handling was recognized and the potential for effectively managing them as an integrated task was discovered. Driven by competitive business environment, the integrated logistics management gained momentum (Daugherty et al., 1996). Because of this recognition, the concept of trade-offs between cost and customer service emerged; this preceded by management practices that could reduce cost while improving the customer service. Later, the integrated distribution trend was broadened to encompass beyond the functional boundaries of a single organization, which lead to the Supply Chain Management (SCM) trend. SCM incorporates both internal and external integrations (Daugherty et al., 1996); internally, supply chain management involves working to achieve a seamless logistics integration with other functional areas of the organization. Externally, the trading partners should plan, execute, and co-ordinate logistical

(37)

performance jointly. Supply chain management is still a popular topic in academia and practice

(Fawcett et al., 2014; Stadtler, 2015)

.

One of the main topics of business conversations in 2000s was the collaboration (Mi et al., 2005). It is referred as the union of two or more companies sharing the responsibility of exchanging common

planning, management, execution, and performance measurement information (Mi et al., 2005). The

main driver of collaboration is to benefit from the synergy between partnering companies and to reach out for levels of performance not achievable individually. There are two types of logistical collaboration: vertical and horizontal. The vertical partnerships are created between firms operating at different levels of the supply chain while in horizontal collaboration partnering companies operate at the same level of the supply chain (Vanovermeire et al., 2014). Scientific research has proven that collaboration implementation has the potential for enhancing the economic and environmental performance of the collaborating companies (Pan et al., 2013; Nagurney et al., 2010; Nagurney, 2009). However, collaboration is inherently challenged by some difficulties. Cruijssen et al. (2007) discuss that several areas such as partner selection, trust, gain division, negotiation, and information and communication technology capabilities trouble the logistical collaboration.

Recently and in line with the advent of Internet and electronic commerce, companies like Amazon.com are rewriting the rules of competition (Kotha, 1998). More precisely, they are offering some customer service initiatives that require highly efficient distribution system such as same day delivery and free same day delivery to prime members in top metropolitan statistics areas (announced in 2015)1_{. Yang (2013) stresses that nowadays the performance dimensions of cost, quality, efficiency}

and customer service level should be all taken into consideration equally.

In addition to today’s highly competitive business environment, there are boundaries on the logistical capabilities, which demonstrate a frontier on distribution performance. Take for instance the limited land available for production and distribution operations (McKinnon, 2009) and expensive and scarce fuel resources (Beamon, 2008).

Physical Internet (Montreuil, 2011; Montreuil, 2012; Montreuil et al., 2013) introduces a novel distribution era; hyperconnected distribution in which businesses exploit the available storage and distribution capacities and technologies belonging to other businesses. According to Montreuil et al.

(38)

(2013), the hyperconnection enables companies to deploy their products close to their geo-markets and eventually offer fast and efficient customer service without the economic burden of high capital investment. The punctuality of the business relationship through hyperconnection, can relax the challenges inherited by long-term commitment involved in logistical collaboration. While a large percentage of operations related to a hyperconnected firm are not performed by them, technologies such as smart PI-container provide visibility and control to Physical Internet users.

Following the evolution of distribution systems discussed here, the next section categorizes the current and prospective distribution systems.

2.4 Distribution system categorization

The eras of atomistic, managerial involvement, outsourcing to 3PLs and integrated distribution, contribute to a system devoted to distributing the products of a single business. It is possible that more than one business operates in this system (e.g. the focal company and 3PLs); however, their involvement is decided and planned by the stakeholders of a single business. Thus, we identify them as dedicated distribution system.

The logical topology of businesses adopting dedicated distribution system is called individualistic. Such businesses invest in storage and distribution facilities individually and create a chain (having one facility for each function such as storage or production) or network (more than one facility for at least one of the functions) dedicated to their own business. They perform deployment, distribution and transportation independently from other businesses; thus, their operations are called private. All the costs related to their distribution system are incurred to the stakeholder(s) of the individual business.

The collaborative distribution era creates the second category called collaborative distribution

system. In such systems, the partnering companies form a coalition of more than one dedicated

system. When a group of partnering businesses create a collaborative distribution system, their logical topology embraces collective goal and behavior. Their physical collaborative topology includes nodes (facilities) and links (operations) which can be exploited by more than one partnering business. However, we distinguish the collaborative physical topology as a shared web to stress that the exploitation of this web is limited to the partnering businesses who formed in the first place. Similarly, the logical topology of the collaborative distribution systems limits the pooling of the operations and facilities to the partnering companies; thus, it is called shared pooling. The costs and savings incurred

(39)

in a collaborative distribution system would be divided among the stakeholder(s) of the partnering businesses (stakeholders).

Lastly, the distribution systems of those businesses practicing distribution by exploiting PI-enabled Mobility Web and Distribution Web are categorized as hyperconnected distribution system. Based on the idea of exploiting DW and MW in a hyperconnected distribution system, the logical topology of hyperconnected distribution systems is called open. Their physical topology would be an open web, since it shapes a network of open distribution networks. The distribution-related operations of hyperconnected distribution systems are recognized as open pooling. The stakeholder(s) of the single business that exploits DW and MW would be responsible for the total distribution costs. Figure 2-2 depicts the differences between entities of each distribution system category.

(40)

Figure 2-3 contrasts the physical topology of the alternative distribution systems using simple schematics. Part a) and b) respectively depict dedicated distribution networks for businesses A and B. Part c) depicts the juxtaposition of the dedicated networks of businesses A and B: two networks overlap, yet there is no connection between them in terms of product flow or shared facility. Part d) depicts the collaborative distribution web for a coalition between businesses A and B. Within the collaborative web, some of the facilities and transportation flows are operated by only one of the two businesses, while several are shared. Part e) depicts the hyperconnected distribution web exploited by businesses A and B, as well as many other PI-certified businesses. The distribution web is depicted as a cloud in rough analogy with cloud storage, currently a strong trend in the Digital Internet. In such a web, the hyperconnected distribution network of each business becomes much more dynamic, evolving to navigate the variations in supply and demand across the territory. Conventional mobility is represented by plain arrows while the dashed (blue) arrows, represent hyperconnected transportation through the Mobility Web. It should be noted that the topologies depicted in Figure 2-3 do not aim to show optimized networks or webs, but to provide a virtual representation of the concepts discussed previously.

(41)

a) Business A, dedicated distribution b) Business B, dedicated distribution c) A& B, disconnected distribution

d) A& B, collaborative distribution e) A& B, Hyperconnected distribution

Figure 2-3. Schematic contrast of dedicated, collaborative and hyperconnected distribution systems (Adapted from Montreuil 2011 and 2012)

(42)

Table 2-1. The evolution of the distribution era

Time Distribution era Distribution system category

< 1960 Atomic

Dedicated 1960s and 1970s Managerial involvement

1980s 3PL 1990s Integration

2000s Collaboration Collaborative

2009 < Hyperconnection Hyperconnected

Based on the research foundations previously presented in this chapter, the next section indicates the questions of this research. They point to the contributions and describe the research methodology.

2.5 Research question and goal

The hyperconnected distribution system is claimed to create potential for drastic economic, environmental and social performance improvements (Montreuil, 2011; Montreuil et al., 2013). The two main questions that have motivated this Ph.D. thesis include:

I. What is the potential economic, social and environmental performance gain by adopting the hyperconnected distribution systems?

II. If the gain is significant, how should a hyperconnected distribution system be planned and managed?

To answer the questions above, the goal of this research is twofold; first, to investigate the potential economic, environmental and social gain enabled by hyperconnected distribution system. Second, it intends to develop conceptual and analytical tools to plan and manage hyperconnected distribution systems.

2.6 Research methodology

An exploratory approach is adopted in this thesis in order to assess the performance of hyperconnected distribution systems through a set of illustrative businesses cases. The economic, environmental and social performance of each business is investigated at existing and prospective distribution systems and the performance gains/loss is analyzed. Our investigation is somehow

(43)

conservative since we suppose that the dedicated and collaborative systems only have access to the current distribution and transportation capacities and technologies, while hyperconnected system is assumed able to exploit both current and PI-enabled Mobility Web and Distribution Web (shown in Figure 2-3). In real life, the exploitation of MW is not limited only to businesses that exploit DW as well (and vice versa). For instance, businesses can exploit open DCs while employing a dedicated transportation system (owned or outsourced transportation). Similarly, they can run owned or leased distribution centers while using MW for transportation.

Multiple approaches have been used in the literature to investigate the performance gain by collaborative logistical systems. Nagurney (2009) quantifies the strategic advantages associated with horizontal collaboration applying a system-optimization perspective. The pre-merger and horizontal-merger transportation networks of two sample firms are optimized and the total transportation costs are compared. Pan et al. (2013) explore the economic and environmental effect of distribution pooling for two French retailers using optimization model at the strategic level. Ghaderi et al. (2012) examined the effect of successful horizontal collaboration among a group of small and medium size enterprises by data collection through quantitative questionnaires. Sarraj et al. (2014) apply multi-agent simulation to assess the efficiency of exploiting PI-enabled mobility web. They develop several transportation and consolidation protocols to enable exploiting mobility web, then simulate both the current and hyperconnected transportation system of two French retailers and their top 106 suppliers.

In this research, optimization is used as the main investigation approach; network modeling helped us to estimate the costs associated with each distribution system at strategic level. Furthermore, the topology of the optimized network/web helped us to assess the associated environmental performance respecting a target social performance. The reason behind selecting optimization is twofold; first, our goal is to reach preliminary results, which can either encourage or discourage hyperconnected distribution systems. For this purpose, optimization at strategic level has been widely applied. Second, other approaches such as simulation, requires more insights about the deployment and distribution practices at operational level. Contrary to the distribution operations of the dedicated and collaborative systems, the hyperconnected distribution operations were unknown for us at the time of performing the investigation.

Hence, in our investigation, first a core modeling approach is developed. Then, it is adjusted to the characteristics of each distribution system to develop three distribution system-driven optimization models. All the optimization models are single-product, single period and deterministic mixed integer programming (MIP) problems. Afterward, a set of illustrative business cases are developed.

(44)

Employing the distribution system-driven MIPs, the distribution network/web of these businesses are designed responding to a set of predetermined policies for customer service. Using a set of Key Performance Indicators (KPIs), the economic and environmental performance of each businesses is obtained and analyzed over each distribution system-service level strategy.

In our quest to reply the second research question, we have developed a conceptual planning framework and a set of analytical tools that can be applied to a future simulation study of hyperconnected distribution assessment.

(45)

(46)

2.7 Research challenges

While addressing the research questions, we faced several challenges mainly raised from the modeling aspect. The characterization of the distribution system user was a demanding task. A large amount of time and effort were invested in this part of research. Particularly, modeling the hyperconnected distribution system was complex; since at the time that this Ph.D. thesis was conducted, the Physical Internet and Logistics Web were in their infancy.

Designing the collaborative distribution webs and dividing the savings between partnering businesses were complicated. It was our intention to follow the core modeling approach and adjust it to the characteristics of each distribution system in order to develop distribution system-driven models. Thus, taking into consideration the effect of transportation pooling in a single-period strategic design model required a meaningful anticipation of the operational level. In addition, developing an efficient method to divide the savings fairly among the partnering businesses is one of the concerns in logistical collaboration modeling. Since the performance gain between multiple distribution systems are to be compared in this research, our approach had to be able of taking into account the potential benefits available for businesses if they operate solely instead of collaboratively.

Furthermore, the modeling and calibration of transportation and distribution costs required data gathering from multiple sources. It was also crucial to adapt the parameter modeling to the characteristics of each distribution system.

The fine-tuning of the solver parameters (here CPLEX), was another challenging aspect, because the optimization models are NP-hard. On the other hand, modeling the operational level details has increased the number of variables, constraints and more importantly the non-zero coefficients. Furthermore, application of piecewise linear cost functions in the objective function intensified the difficulty of solving the optimization problems. These functions linearize the transportation and DC opening and warehousing costs which are modeled as non-linear parameters. At last, the analysis of large amount of numerical results obtained by our experimentation for several businesses and multiple customer service targets required precision to derive meaningful conclusions.

By overcoming the research challenges, we provided answers to the research questions and contributed to the network modeling, optimization and Physical Internet research areas as listed in the following section.

(47)

2.8 Research contributions

By overcoming the previously discussed challenges, this thesis contributes to (a) Physical Internet and (b) Distribution/Supply Chain Modeling and Optimization fields.

a) Physical Internet:

 Systematic performance investigation

 Hyperconnected distribution planning framework

b) Distribution/Supply Chain Modeling approach and Optimization:

 Core distribution network design modeling approach

 Three distribution system-driven models

 Transportation modeling

 Formulation of stochastic hyperconnected distribution web design model

 Development of nonlinear functions

 Development of coalition savings division approach

 The business case development and parameter estimation

 Environmental performance assessment approach

(48)

Table 2-2. Summary of the main research blocks

Research motivation

 Investigate the potential gain by adopting hyperconnected distribution system

 Provide conceptual and analytical tools to plan and manage hyperconnected distribution systems

Research challenges

 Characterization of the user level (anticipation of the operational level)  Collaborative distribution web design and savings division approach  Parameter modeling for status-quo and hyperconnected distribution system  Solve the large MIPs obtained efficiently

 Interpret and analyze the large information provided by MIP solutions

Research contributions

 Physical Internet field

 Distribution/Supply Chain Modeling and Optimization field

Research results/vision

 The magnitude of performance gain by exploiting the hyperconnected distribution system  Conditions and assumption

(49)

3

Chapter 3

(50)

3.1 Introduction

The goal of developing optimization models in this chapter is to design the distribution network of a set of business cases by adopting the principles of the dedicated, collaborative and hyperconnected distribution systems and eventually, assess their economic and environmental performance.

Distribution Network Design Problems involve strategic decisions on the number, location, capacity and mission of a set of distribution centers, their allocation to a set of demanding customers while optimizing the performance of the entire network by either minimizing the total associated costs or maximizing the profit (Klibi et al., 2010). Model formulations and solution algorithms which address this issue vary widely in terms of fundamental assumptions, mathematical formulation and solution approach.

In this section, first the literature of the discrete facility location problem, particularly the Mixed Integer Programming category to which our model belongs, is reviewed. Then, the context of the problem on hand is described and our core modeling approach is provided. Three system driven models are derived from the proposed core modeling approach.

(51)

3.2 Literature review

The strategic decision to design the distribution network, mainly involves decisions regarding the location of distribution and storage facilities. Facility location is a well-established field within Operations Research. More than a century ago, Weber pioneered the Location Theory. He searched for the location of a single facility to be served by two sources and to supply a single market while minimizing the total transportation cost (English translation of the original work in 1909 in German: Weber and Friedrich, 1929). In 1960s, Hakimi extended the location theory by developing the p-median problem, where p facilities have to be located on a graph such that the sum of distances between the nodes of the graph and the nearest facility is minimized (Hakimi 1964 and 1965). Since then, various forms of location models have been created.

Facility location problems can be categorized as continuous and discrete. The solution space of the first category is continuous; thus, facilities can be located on any point in the solution plane. Moreover, the distance between points is measured while solving the model applying a suitable metric (such as Euclidean distance metric). Discrete facility locations however, locate facilities among a predetermined set of potential sites. Klose and Drexl (2005) divide this category of models into Network Location models and Mixed-Integer Programming models (MIP). The former includes models inspired by p-median to locate a predetermined number of facilities on a graph while minimizing the total distance in the network. While MIP models go beyond simply minimizing the total distance by taking into account various decisions such as production/storage, capacity acquisition, customer service level, transportation mode selection and integration of tactical and operational decisions in the strategic facility location problem.

Several classification and revision of facility location models exist in the literature. Owen and Daskin (1998) and Drezner and Hamacher (2002) review the early works in this field. Furthermore, Geoffrion and Powers (1995) and Shapiro et al. (1993) discuss the evolution of the strategic supply chain design models. For more recent reviews, the reader is referred to Şahin and Süral (2007), Klibi et al. (2010) and Arabani and Farahani (2012).

Since our core modeling approach will be aggregating operations on the planning horizon without loss of generality, the focus of our literature review is given to the single period models. Klibi et al. (2015) discuss that when the demand behavior is more stable, a single period model can provide near optimal solution.

(52)

The single-period network design MIPs in the literature, mostly incorporate cost minimization objective function. However, these models vary in terms of fundamental assumptions, mathematical formulation and solution approach. Jayaraman and Ross (2003), Geoffrion and Grave (1974) introduce single-echelon multi-product distribution network design models where, the costs are modeled as linear parameters and customers are single-sourced. The former is solved by Simulated Annealing and the latter, by Bender Decomposition. Syarif et al. (2002) develop a single-echelon single-product production and distribution network design problem with linear costs and multi-sourcing. They have solved the problem with a spanning tree-based genetic algorithm.

Some contributions involve more precision in modeling cost parameters. Amiri (2006) has developed a single-product multi-sourcing problem with non-linear DC opening costs. Baumgartner et al. (2012) and Fleischmann (1993) modeled warehousing and transportation cost by taking into account the economies of scale. Shen and Daskin (2005) modeled the inventory and safety stock cost as nonlinear functions. Tsao and Lu (2012), take into account the impact of both economies of distance and scale on the transportation cost. The latter does not model these nonlinearities simultaneously; the economies of scale is only associated with the inbound transportation and the economies of distance, solely with the outbound (DC-to-Market zone flow).

Several single-period contributions have introduced transportation modeling. In Sadjady and Davoudpour (2012), the transportation mode selection is invoked by the monetary value of the overall lead time in addition to the cost. The latter has also introduced a capacitated version of the model where transportation modes are attributed a total volume/weight-based capacity. However, the application of the model is uncapacitated. The single-period network design model developed by Eskigun et al. (2005) pre-assigns rail transportation to inbound links and road transportation, to the outbound. However, shipping through a DC or directly to the customer is influenced by the monetary value of the lead time. In Cordeau et al. (2006), the space required by each product is constrained to the available capacity offered by each mode on each network link. However, in the numerical application the link capacity is not binding the flow (serves as a “big M”). Thus, on each transportation link, the cheapest mode is selected. Lapierre et al. (2004) involves direct and indirect transportations (through a transportation/consolidation center) costs. The characteristics of transportation cost function for various options including truckload, less than truckload and parcel is detailed. However, no modeling effort is devoted to the selection of the transportation option in the mathematical model.

(53)

The customer service has been modeled in several single-period models. Amrani et al. (2011) ensure the customer service quality by applying an upper bound on the permitted distance between DC and customer. Eskigun et al. (2005) model the customer service by embedding the monetary value of the total lead time in the minimization objective function. In Ambrosino and Scutella (2005) customers are classified based on their demand level. The customer service is modeled by ensuring a minimum inventory level at each DC influenced by the classification of customers supplied by it to guarantee product availability especially for the most valued customers.

Only a few models in this category consider two-echelon storage and distribution settings. Sadjady and Davoudpour (2012) model storage at plant and warehouse and Ambrosino and Scutella (2005), in both regional and central DCs.

In addition to designing the physical structure, network design models have been applied to investigate the impact of managerial insights on overall performance. Particularly, single-period deterministic models serve this purpose, since they can be fairly easy to solve. For instance, Nagurney (2009) explores the strategic advantages of merging supply chain networks by comparing the total cost of two firms before and after being merged. The total cost however, has been calculated by modeling the economic activities of firms with a single-period network design model.

The ease of solving single-period models is unwantedly accompanied by lower accuracy of anticipating the user response to design decisions. Klibi and Martel (2015) argue that the transportation quantity in these models, can easily fall into bulky transportation modes such as truckload, which incur lower cost compared to less than truckload. To our knowledge, anticipating transportation operations at operational level is poorly addressed in single-period network design problems to date. Regardless of some modeling efforts and usage of real world data, often, the model simply selects the cheapest transportation mode available. Moreover, estimating the costs associated with them is challenging. Since distribution-related costs manifest nonlinear behavior, the accuracy of their estimation strongly depends on the quality of user response anticipation.

Table 3-1 summarizes our review of several single-period Distribution Network Design models. The intention of providing this table is to highlight the diversity of modeling assumptions, approaches and solution algorithms within this group of models for the rather than offering a taxonomy or literature classification. Besides, this table can indicate the current gap in literature and underline the value contributions of our model.

(54)

Table 3-1. The modeling aspects incorporated in a selected set of single-period MIP distribution network design models Selected Reference Modeling Aspect Geoffrion and Grave (1974) Fleischmann (1993) Cole (1995) Tsiakis et al. (2001) Nozick and Turnquist (2001) Syarif et al. (2002) Jayaraman and Ross (2003) Eskigun et al. (2005) Ambrosino and Scutella (2005) Shen and Daskin (2005) Objective function (Minimize cost/ Maximize profit)

Min Min Min Min Min Min Min Min Min Min

Number of product (Single/Multiple) M M M M S S M S (6) _{S and M}(9)_S Type of facility to locate (DC/Plant/Hub)

DC DC & Hub DC DC DC Plant & _DC₍₄₎ DC & Hub (4) _DC _{DC & Hub} _DC

Facility capacity (Capacitated/ Uncapacitated) C C C C U C C C C U Number of facility echelon to locate 1 2 (DC) & _1(hub) 2 2 1 1 1 1 2 1 Demand behavior (Deterministic/ Stochastic)

D S S _providedBoth are S D D D D S

Customer sourcing (Single/Multiple) S S S (2) _{S S}_M_S (5) _S _{M S} Customer service X (1) (1) X (3) X X (7) (10) (11)

Cost modeling All linear

Nonlinear DC and transportation cost Nonlinear DC cost Nonlinear transportation cost Nonlinear

DC cost All linear

Nonlinear transportation cost Nonlinear transportation cost All linear Nonlinear inventory and transportation cost Transportation mode selection X X X X X X X (8) X X Solution

approach decomposition Bender

Iterative linearization

technique

Commercial

Solver Commercial Solver heuristic Hybrid algorithm Genetic Annealing Simulated Lagrangian heuristic Commercial Solver

Genetic